Running OpenClaw locally is great for testing, but production is where complexity appears: uptime, secrets, Docker, VPS, SSL, and channel reliability.
This tutorial shows how to launch and operate OpenClaw using AgentClaw.app so you can focus on your agent logic instead of infrastructure.
Why AgentClaw.app?
With AgentClaw, you can:
- Deploy OpenClaw without manual server setup
- Manage lifecycle from a dashboard (start/stop/restart/delete)
- Configure Anthropic/OpenAI models via UI
- Add Telegram/Discord integrations
- Store sensitive env vars encrypted
- Access your instance via HTTPS subdomain
What You’ll Build
By the end of this guide, you’ll have:
- A cloud-hosted OpenClaw instance
- Model provider and API key configured
- Optional channel integration (Telegram/Discord)
- A repeatable operations flow for updates and troubleshooting
Prerequisites
Before starting, prepare:
- An AgentClaw.app account
- One model provider API key:
- Anthropic key or
- OpenAI key
- Optional:
- Telegram bot token
- Discord bot token
Step 1 — Create Your Instance
- Open your AgentClaw dashboard.
- Navigate to Instances → New Instance.
- Fill:
- Name (friendly label)
- Subdomain (public URL prefix)
- Choose initial AI provider/model.
- Add API key and optional channel tokens.
- Submit.
You now have an instance entry ready for provisioning and runtime actions.
Step 2 — Configure AI Provider & Model
Open your instance detail and configure:
- Provider: Anthropic (Claude) or OpenAI (GPT)
- Model: pick from list or enter custom provider/model id
- API key: stored as encrypted environment variable
If you change provider or model later, run Restart so the runtime picks up the new config.
Step 3 — Configure Channels (Optional)
Go to Config → Channels.
Telegram
- Enable Telegram
- Add
TELEGRAM_BOT_TOKEN - Set DM policy (pairing / allowlist / open / disabled)
- Approve pending pairings if required
Discord
- Add
DISCORD_BOT_TOKEN - Follow the UI prompts for your current integration flow
Start with one channel first, validate it, then add others.
Step 4 — Start the Instance
From the instance page, click Start.
Typical platform flow:
- Assign worker node
- Write secure runtime env
- Prepare OpenClaw workspace/runtime
- Launch container
- Mark status as RUNNING
When successful, your instance is available via HTTPS subdomain.
Step 5 — Validate Deployment
Use this checklist:
- Instance status = RUNNING
- URL opens over HTTPS
- Model calls succeed (no auth/model errors)
- Channel status is healthy
- Logs show normal startup (no crash loop)
If behavior doesn’t reflect recent config changes, run Restart.
Step 6 — Daily Operations
Use this lifecycle pattern:
- Start: bring agent online
- Stop: maintenance/off-hours control
- Restart: apply model/env/channel updates
- Delete: decommission instance
Keep config changes centralized in AgentClaw instead of editing scattered server files.
Troubleshooting Quick Guide
Instance fails to start
- Verify API key validity
- Confirm model id is available for your account
- Check instance logs/diagnostics
- Fix config and Restart
Channel is not responding
- Re-check token value
- Validate bot permissions on the platform side
- Confirm DM/pairing policy settings
Config change not applied
- Run Restart (required for some runtime changes)
Best Practices
- Use one instance per business use case (support, sales, ops, etc.)
- Keep secrets only in AgentClaw-managed encrypted config
- Track prompt/model changes in a changelog
- Roll out model switches gradually and monitor logs before broad rollout
Final Thoughts
AgentClaw.app helps you move from “it works on my machine” to “it runs reliably in production” for OpenClaw.
If your priority is shipping AI agents quickly without managing Docker + VPS internals, this is a practical and scalable workflow.
Top comments (0)